draft: add cumsum,cumproduct,cummin,cummax (#293)

* add cumsum,cumproduct,cummin,cummax

* reformat

* reformat

* rename

* adding new cumulative specs and cumprod,cumsum tests

* updating cummax and cummin testing

* fix def cummin

* update cum-process tesing with nan values

Author: katharinastarzer21

openeo_processes_dask/process_implementations/math.py

@@ -26,6 +26,10 @@
     "add",
     "_sum",
     "_min",
+    "cumsum",
+    "cumproduct",
+    "cummin",
+    "cummax",
     "_max",
     "median",
     "mean",
@@ -117,6 +121,58 @@ def _min(data, ignore_nodata=True, axis=None, keepdims=False):
         return np.min(data, axis=axis, keepdims=keepdims)
 
 
+def cumsum(data, ignore_nodata=True, axis=None):
+    nan_mask = np.isnan(data)
+
+    if ignore_nodata:
+        result = np.nancumsum(data, axis=axis)
+    else:
+        result = np.cumsum(data, axis=axis)
+
+    result[nan_mask] = np.nan
+    return result
+
+
+def cumproduct(data, ignore_nodata=True, axis=None):
+    nan_mask = np.isnan(data)
+
+    if ignore_nodata:
+        result = np.nancumprod(data, axis=axis)
+    else:
+        result = np.cumprod(data, axis=axis)
+
+    result[nan_mask] = np.nan
+    return result
+
+
+def cummin(data, ignore_nodata=True, axis=None):
+    data = np.array(data)
+    nan_mask = np.isnan(data)
+
+    if ignore_nodata:
+        data_filled = np.where(nan_mask, np.inf, data)
+        result = np.minimum.accumulate(data_filled, axis=axis)
+    else:
+        result = np.minimum.accumulate(data, axis=axis)
+
+    result[nan_mask] = np.nan
+    return result
+
+
+def cummax(data, ignore_nodata=True, axis=None):
+    data = np.array(data)
+    nan_mask = np.isnan(data)
+
+    if ignore_nodata:
+        data_filled = np.where(nan_mask, -np.inf, data)
+        result = np.maximum.accumulate(data_filled, axis=axis)
+    else:
+        result = np.maximum.accumulate(data, axis=axis)
+
+    result[nan_mask] = np.nan
+    return result
+
+
 def _max(data, ignore_nodata=True, axis=None, keepdims=False):
     if ignore_nodata:
         return np.nanmax(data, axis=axis, keepdims=keepdims)

tests/test_apply.py

@@ -347,3 +347,193 @@ def test_apply_kernel(temporal_interval, bounding_box, random_raster_data):
     )
 
     xr.testing.assert_equal(output_cube, input_cube)
+
+
+@pytest.mark.parametrize("size", [(6, 5, 30, 4)])
+@pytest.mark.parametrize("dtype", [np.float32])
+def test_apply_dimension_cumsum_process(
+    temporal_interval, bounding_box, random_raster_data, process_registry
+):
+    input_cube = create_fake_rastercube(
+        data=random_raster_data,
+        spatial_extent=bounding_box,
+        temporal_extent=temporal_interval,
+        bands=["B02", "B03", "B04", "B08"],
+        backend="dask",
+    )
+
+    _process_cumsum = partial(
+        process_registry["cumsum"].implementation,
+        data=ParameterReference(from_parameter="data"),
+    )
+
+    output_cube_cumsum = apply_dimension(
+        data=input_cube,
+        process=_process_cumsum,
+        dimension="t",
+    ).compute()
+
+    original_abs_sum = np.sum(np.abs(input_cube.data))
+
+    cumsum_total = np.sum(np.abs(output_cube_cumsum.data))
+
+    assert cumsum_total >= original_abs_sum
+
+    input_cube.data[:, :, 15, :] = np.nan
+
+    _process_cumsum_with_nan = partial(
+        process_registry["cumsum"].implementation,
+        data=ParameterReference(from_parameter="data"),
+        ignore_nodata=False,
+    )
+
+    output_cube_cumsum_with_nan = apply_dimension(
+        data=input_cube,
+        process=_process_cumsum_with_nan,
+        dimension="t",
+    ).compute()
+
+    assert np.isnan(output_cube_cumsum_with_nan[0, 0, 20, 0].values)
+
+
+@pytest.mark.parametrize("size", [(6, 5, 30, 4)])
+@pytest.mark.parametrize("dtype", [np.float32])
+def test_apply_dimension_cumproduct_process(
+    temporal_interval, bounding_box, random_raster_data, process_registry
+):
+    input_cube = create_fake_rastercube(
+        data=random_raster_data,
+        spatial_extent=bounding_box,
+        temporal_extent=temporal_interval,
+        bands=["B02", "B03", "B04", "B08"],
+        backend="dask",
+    )
+
+    _process_cumsum = partial(
+        process_registry["cumproduct"].implementation,
+        data=ParameterReference(from_parameter="data"),
+    )
+
+    output_cube_cumprod = apply_dimension(
+        data=input_cube,
+        process=_process_cumsum,
+        dimension="t",
+    ).compute()
+
+    original_data = np.abs(input_cube.data)
+    original_data[np.isnan(original_data)] = 0
+    original_abs_prod = np.sum(original_data)
+
+    cumprod_data = np.abs(output_cube_cumprod.data)
+    cumprod_data[np.isnan(cumprod_data)] = 0
+    cumprod_total = np.sum(cumprod_data)
+
+    assert cumprod_total >= original_abs_prod
+
+    input_cube.data[:, :, 15, :] = np.nan
+
+    _process_cumprod_with_nan = partial(
+        process_registry["cumproduct"].implementation,
+        data=ParameterReference(from_parameter="data"),
+        ignore_nodata=False,
+    )
+
+    output_cube_cumprod_with_nan = apply_dimension(
+        data=input_cube,
+        process=_process_cumprod_with_nan,
+        dimension="t",
+    ).compute()
+
+    assert np.isnan(output_cube_cumprod_with_nan[0, 0, 20, 0].values)
+
+
+@pytest.mark.parametrize("size", [(6, 5, 30, 4)])
+@pytest.mark.parametrize("dtype", [np.float32])
+def test_apply_dimension_cummax_process(
+    temporal_interval, bounding_box, random_raster_data, process_registry
+):
+    input_cube = create_fake_rastercube(
+        data=random_raster_data,
+        spatial_extent=bounding_box,
+        temporal_extent=temporal_interval,
+        bands=["B02", "B03", "B04", "B08"],
+        backend="dask",
+    )
+
+    _process_cummax = partial(
+        process_registry["cummax"].implementation,
+        data=ParameterReference(from_parameter="data"),
+    )
+
+    output_cube_cummax = apply_dimension(
+        data=input_cube,
+        process=_process_cummax,
+        dimension="t",
+    ).compute()
+
+    original_abs_max = np.max(input_cube.data, axis=0)
+    cummax_total = np.max(output_cube_cummax.data, axis=0)
+
+    assert np.all(cummax_total >= original_abs_max)
+
+    input_cube.data[:, :, 15, :] = np.nan
+
+    _process_cummax_with_nan = partial(
+        process_registry["cummax"].implementation,
+        data=ParameterReference(from_parameter="data"),
+        ignore_nodata=False,
+    )
+
+    output_cube_cummax_with_nan = apply_dimension(
+        data=input_cube,
+        process=_process_cummax_with_nan,
+        dimension="t",
+    ).compute()
+
+    assert np.isnan(output_cube_cummax_with_nan[0, 0, 16, 0].values)
+
+
+@pytest.mark.parametrize("size", [(6, 5, 30, 4)])
+@pytest.mark.parametrize("dtype", [np.float32])
+def test_apply_dimension_cummin_process(
+    temporal_interval, bounding_box, random_raster_data, process_registry
+):
+    input_cube = create_fake_rastercube(
+        data=random_raster_data,
+        spatial_extent=bounding_box,
+        temporal_extent=temporal_interval,
+        bands=["B02", "B03", "B04", "B08"],
+        backend="dask",
+    )
+
+    _process_cummin = partial(
+        process_registry["cummin"].implementation,
+        data=ParameterReference(from_parameter="data"),
+    )
+
+    output_cube_cummin = apply_dimension(
+        data=input_cube,
+        process=_process_cummin,
+        dimension="t",
+    ).compute()
+
+    original_abs_min = np.min(input_cube.data, axis=0)
+    cummin_total = np.min(output_cube_cummin.data, axis=0)
+
+    assert np.all(cummin_total <= original_abs_min)
+
+    input_cube.data[:, :, 15, :] = np.nan
+
+    _process_cummin_with_nan = partial(
+        process_registry["cummin"].implementation,
+        data=ParameterReference(from_parameter="data"),
+        ignore_nodata=False,
+    )
+
+    output_cube_cummin_with_nan = apply_dimension(
+        data=input_cube,
+        process=_process_cummin_with_nan,
+        dimension="t",
+    ).compute()
+
+    assert np.isnan(output_cube_cummin_with_nan[0, 0, 16, 0].values)

tests/test_math.py

@@ -118,3 +118,59 @@ def test_extrema():
     dask_array = da.from_array(np.array(array_list))
     result = extrema(dask_array, ignore_nodata=True, axis=0, keepdims=False)
     assert np.array_equal(result_np, result.compute())
+
+
+def test_cumproduct():
+    array_list = [1, 2, 3, np.nan, 3, 1]
+    result_np = [1, 2, 6, np.nan, 18, 18]
+
+    result = cumproduct(array_list)
+    assert np.array_equal(result_np, result, equal_nan=True)
+
+    array_list = [1, 2, 3, np.nan, 3, 1]
+    result_np = [1, 2, 6, np.nan, np.nan, np.nan]
+
+    result = cumproduct(array_list, ignore_nodata=False)
+    assert np.array_equal(result_np, result, equal_nan=True)
+
+
+def test_cumsum():
+    array_list = [1, 3, np.nan, 3, 1]
+    result_np = [1, 4, np.nan, 7, 8]
+
+    result = cumsum(array_list)
+    assert np.array_equal(result_np, result, equal_nan=True)
+
+    array_list = [1, 3, np.nan, 3, 1]
+    result_np = [1, 4, np.nan, np.nan, np.nan]
+
+    result = cumsum(array_list, ignore_nodata=False)
+    assert np.array_equal(result_np, result, equal_nan=True)
+
+
+def test_cummin():
+    array_list = [5, 3, np.nan, 1, 5]
+    result_np = [5, 3, np.nan, 1, 1]
+
+    result = cummin(array_list)
+    assert np.array_equal(result_np, result, equal_nan=True)
+
+    array_list = [1, 3, np.nan, 3, 1]
+    result_np = [1, 1, np.nan, np.nan, np.nan]
+
+    result = cummin(array_list, ignore_nodata=False)
+    assert np.array_equal(result_np, result, equal_nan=True)
+
+
+def test_cummax():
+    array_list = [1, 3, np.nan, 5, 1]
+    result_np = [1, 3, np.nan, 5, 5]
+
+    result = cummax(array_list)
+    assert np.array_equal(result_np, result, equal_nan=True)
+
+    array_list = [1, 3, np.nan, 3, 1]
+    result_np = [1, 3, np.nan, np.nan, np.nan]
+
+    result = cummax(array_list, ignore_nodata=False)
+    assert np.array_equal(result_np, result, equal_nan=True)